In the Magazine Tech & Science

AI Promises Life-Changing Alzheimer's Drug Breakthrough

HOR_Maney_Alzheimers_01_687788323
Human brain and circuit board, illustration. CHRISTIAN LAGEREK/SCIENCE PHOTO LIBRARY/Getty

If Verge Genomics co-founder Alice Zhang is right, the kind of technology that allows you to search the web for “Japanese baseball jerseys” and find a $49.99 Hokkaido Nippon-Ham Fighters shirt will help discover a cure for Alzheimer’s.

And that’s just the beginning. Her company could make traditional drug research in a lab seem more outmoded than an Amish wagon on an eight-lane highway. “Verge moves drug discovery from the lab to the computer,” Zhang says. Today, most new Big Pharma drugs have gone through a painstaking process of lab research and trials that went on for 10 to 15 years and cost as much as $2 billion. Verge seems to be showing how to take that analog process and make it digital. And we all know what happens when anything goes from analog to digital: costs dive and supply shoots up. People used to buy a vinyl album for $12 but now can just stream it on Spotify for nearly nothing. In Verge Genomics, you can see the promise of a similar transformation in pharmaceuticals. We’d get a multiplicity of new drugs flowing onto the market for a fraction of the ridiculous prices we now pay.

Verge has been getting a good deal of attention, in part because of Zhang—a 28-year-old Ph.D. dropout female CEO who vows to find treatments for so-far-incurable neurological diseases by using artificial intelligence. For the media, that’s a pretty tasty stew. Her story goes like this: She was at the University of California, Los Angeles, going for a neuroscience Ph.D. and feeling frustrated at the pace of drug development. As she tells me, drug researchers usually zero in on one gene at a time, looking for a key to turning off a disease like Alzheimer’s or Parkinson’s. Yet such diseases are typically caused by interactions within a complex network of genes, and few tools exist for understanding those networks. For her Ph.D. program, Zhang wrote software to find such networks of problematic genes—an approach inspired by the algorithm behind Google’s search engine, which looks for billions of connections between keywords, websites and user activity to find just the right results.

Zhang initially used the technology to look for networks of genes that might help nerves regenerate after an injury. Her first success at UCLA was finding a compound that helped mice recover the use of their legs after they were crushed. The mice got better four times faster than through natural healing processes. (I didn’t want to ask how the mice’s legs were crushed. Tiny steamrollers driven by mobster rats?) “Others had tried thousands of drugs and got nothing,” Zhang says. With her software, she quickly got to a drug that worked.

Zhang concluded that writing a Ph.D. thesis a handful of scientists might read wasn’t going to have the impact she wanted. She had grown up with an activist mentality—her father was part of the Democracy Wall movement in China in the 1970s and fled to the U.S. in the 1980s. She felt she had to commercialize her work. “I left three months shy of graduating, much to the anger of my parents,” she says. She co-founded Verge with biomedical engineer Jason Chen, and the duo won a slot at the Y Combinator startup accelerator in Silicon Valley. In 2015, Verge took in $4 million in venture capital. The company hired a team of neuroscientists and computer scientists, threw them into an office and had them build sophisticated AI to map interactions among genes involved in a neurological disease. Once the networks are mapped, Verge’s software looks for known drugs that can affect all the genes in the network at once to essentially turn off the disease. The technology can look for millions of possible answers in the time it would take a human to set up a single lab experiment.

As with any AI, the key to the Verge’s technology is an enormous amount of data that the AI can learn from. Such data has exploded in recent years, thanks to cheap and easy genetic testing. The startup Color Genomics, for instance, offers a genetic test for certain kinds of cancer risks for $99. In 2001, sequencing one genome would’ve cost $100 million. To get huge genetic data sets to feed its AI, Verge forged one partnership in September to obtain data from Columbia University and three other universities, and another in November with the National Institutes of Health, Scripps Research Institute and the Dresden University of Technology.

Zhang believes Verge will have a drug for amyotrophic lateral sclerosis in clinical trials within five years. Today there is no cure for ALS (aka Lou Gehrig’s disease). “Once we have our first success in a disease, what we do will be very scalable,” Zhang says. As Verge and other companies hone AI-based approaches to drug discovery, new cures for neurological conditions—including Alzheimer’s—could come quickly. “Our software platform will be able to discover not one drug but dozens of billion-dollar drugs,” Zhang adds.

Verge isn’t the only company to see this trend toward AI drug discovery. Atomwise, founded by AI researchers from the University of Toronto, has raised $6 million and is collaborating with Stanford University, IBM, Merck and two dozen other companies and research labs. Atomwise’s technology uses machine learning to study how potential drug molecules interact and bind with target molecules in the body, and promises to make predictions about drug compounds before physically testing them in labs. That could greatly speed discovery of effective drugs. IBM researchers have been developing AI tools that can look for patterns in the side effects of existing drugs and make predictions about what other conditions a drug might treat. A company called ID Genomics is using machine learning to predict which antibiotics will be most effective on different bugs, helping patients get better faster, with fewer pills.

Zhang foresees the rise of a new generation of pharmaceutical companies built on software and computation. Digitizing drug discovery could save an average of $330 million per approved drug, according to a recent Morgan Stanley analysis. The major drug companies will be forced to join this wave, notes Ricky Goldwasser, co-author of the report—or they’ll wind up as creaky as a company like Sears, which never made a transition to the digital era.

Falling drug prices would be a huge win for the public, of course. But it’s a much bigger win if AI can help us crack neurological diseases like Alzheimer’s, which afflicts 5.5 million Americans. That would be a much better outcome from search technology than just hoping Google will someday answer the query “Where are my keys?”

Join the Discussion